A customer used to type "best plumber near me" into Google. Increasingly they ask an AI assistant the same thing, in a full sentence, and get one tidy answer instead of ten blue links.
The question every owner then asks is the same: "Does it mention me?"
Nobody can promise you that it will. But you can find out whether it does in ten minutes, and you can fix the parts that are within your control. Here is how.
What is my take on AI search?
My view: treat AI search as search with a new front door, not as a new religion. People selling secret tricks for getting named by assistants are guessing, and so is everyone else, because the companies behind the assistants do not publish the rules. What I am confident about is the boring part: a site that can be read, says the same thing everywhere and answers real questions is easier to quote than one that does not. That is an opinion with Google's own guidance behind it, not a promise.
How do AI assistants decide which businesses to mention?
Honestly, nobody outside the companies knows the full recipe, and it changes. What is public is thin. Google's own page on AI features and your website says that the basics of good search still apply: pages that can be crawled, indexed, and that are genuinely helpful.
So I treat AI search as search with a new front door. If your information is clear, consistent and easy to read for a machine, you are easier to quote. If it is buried in images, hidden behind pop-ups or contradicted from one site to the next, you are harder to quote. That is a practical way to think, even where the rules are fuzzy.
What is the 10-minute test?
You will need a notepad and an assistant or two (ChatGPT, Gemini, Perplexity, whichever your customers use).
- Write down five questions a customer would ask before choosing you. Not "tell me about Acme Ltd". Questions like "who builds custom decks in Cape Town?" or "how much does a small-business website cost in South Africa?".
- Ask each assistant all five. Open a fresh chat each time so earlier answers do not colour the next.
- For each answer, write down three things: were you named, who was named instead, and what source did it cite?
- Repeat in a week. These answers move. One test is a snapshot, not a verdict.
The part people skip is step three. The names of the competitors who do appear, and the pages the assistant quotes, are a map of what it considers a good answer.
What can I actually fix?
Here are the parts that are in your hands.
Can the assistants even reach my site?
Check your robots.txt file. If a blanket rule blocks crawlers, you have made yourself invisible on purpose. It happens more than you would think, usually left over from when the site was being built.
Does my page make sense without JavaScript?
Some crawlers do not run the scripts that draw a page. If your key text only appears after a script runs, they may see an empty page. Open your page with scripts turned off and see whether the important words are still there.
Is my information consistent?
Business name, location, services and prices should match on your site and everywhere else you appear. When sources disagree, a careful system has less reason to trust any of them.
Do I answer questions directly?
Assistants answer questions, so pages that state a plain answer near the top, and then explain, are easier to draw from. That is why this post, like all of mine, opens each section with the answer.
Is my data labelled?
Structured data is hidden labelling that says "this is a business, this is a service, these are the opening hours". Google explains it in its introduction to structured data. It is written for machines and it is cheap to add.
Do I have an llms.txt file?
This one is optional and unproven. The llms.txt proposal suggests a plain-text summary of your site that is easy for AI tools to read. No one has promised me that any assistant reads it, so I treat it as a free extra and not a strategy. My own site has one.
What results did this approach get me?
I will keep this honest. On the online store I look after, the AI-search readiness score went from C+ to A+ after the clean-up. A readiness score is not the same thing as being quoted by an assistant, and I do not claim it is. It means the site stopped giving machines reasons to skip it.
The same work also produced something I can measure: 4 to 5 quote requests a week, all organic. The pattern holds. Clear pages, correct data, a crawlable site. It is not glamorous. It does not require a secret.
What should I do first?
If you only have an afternoon:
- Run the 10-minute test and write down the results.
- Check robots.txt and a page with scripts off.
- Make sure your name, address and services match everywhere.
- Add a plain question-and-answer section to your most important page.
If you would like a hand reading the results, I am glad to talk it through, and the case files show how I handled it on a real store.
Frequently asked questions
Can I pay to appear in an AI assistant's answer?
Some platforms are testing ads, but the rules are changing quickly. I would not build a plan on it yet.
Does AI search replace Google?
Not today. People use both. The practical approach is to make your site good for both, since the fixes overlap heavily.
Is it worth optimising for AI search if I get few visits from it?
The fixes are the same ones that help normal search: clear pages, correct data, a crawlable site. You are not choosing between them.
How often should I run the test?
Monthly is plenty. Keep the notes so you can see movement over time.
